This repository serves as a curated collection of research papers and projects focused on Physics-Based Deep Learning (PBDL). It targets researchers and engineers interested in combining physical modeling with deep learning techniques, particularly for solving forward and inverse problems in physics, with a strong emphasis on fluid dynamics. The primary benefit is providing a comprehensive overview and links to state-of-the-art methods in this rapidly evolving field.

How It Works

The repository categorizes PBDL approaches based on the integration level of physical models into the deep learning process: data-driven methods, loss-term encoding of physical dynamics, and interleaved simulations where physical simulators are fully differentiable and integrated with neural networks. This classification highlights how different methods leverage differentiable physics for tight integration with numerical simulations, enabling more accurate and efficient solutions for complex physical problems.

Quick Start & Requirements

This repository is a collection of links to papers and projects, not a runnable codebase. Specific requirements will vary per linked project.

Highlighted Details

• Extensive coverage of fluid dynamics problems, including Navier-Stokes equations, turbulence modeling, and flow simulations.
• Categorization of PBDL methods by integration level (data-driven, loss-terms, interleaved) and problem type (forward/inverse).
• Links to a comprehensive PBDL book and numerous associated GitHub repositories for specific projects.
• Includes benchmarks and datasets like APEBench and WeatherBench for evaluating PBDL models.

Maintenance & Community

The repository is maintained by the I15 lab at TUM (Technical University of Munich). Contributions and suggestions for overlooked papers can be sent to i15ge@cs.tum.de. The project homepage is available at https://ge.in.tum.de/.

Licensing & Compatibility

The repository itself is a collection of links and does not have a specific license. The licenses of the linked projects vary and must be checked individually.

Limitations & Caveats

This is a curated list of links, not a unified framework or library. Users must navigate to individual projects for code, setup instructions, and specific dependencies. The field is rapidly advancing, so the list may not be exhaustive or perfectly up-to-date.